The W3C organization has released a first draft of a proposed neural network API in the browser. The Web Neural Network API should enable the browser to be used for machine learning.
A first draft of the proposal is now available on the site of the standardization organisation. The document was prepared by the Web Machine Learning Working Group, which was adopted last year. That group includes Google, Microsoft and Intel.
The proposed standard should be called WebNN, or Web Neural Network. It is an api that makes it possible to use neural networks from within the browser by using hardware acceleration APIs such as WebGPU or WebGL. The W3C mentions object or facial recognition as possible applications, but also live captioning, translations or noise suppression. Such applications can then be run client-side.
The proposal contains various proposals regarding the security and privacy conditions for the API. It should be turned off by default so that not every website administrator can just use it without permissions. The API also prescribes that images and videos used for calculations can only run in the sandbox.
